That's a good point. Technically you're correct about the blocksize limit not being part of the protocol, of course.
Though if everyone is running a client with a different blocksize, the network will become a clusterfuck.
No. The miners -- or rather 51% of them -- would then determine the max block size. No clusterfuck. Longest chain rules.
And they've chosen to keep Satoshi's 1MB block size in place all these years. Anyway, how would 51% of miners be able to coordinate on a new block size? There is no Schelling Point above 1MB, so you'd end up with a bunch of forks.
The immutability of the 1MB blocksize still stands if we throw out the word "protocol". Satoshi knew that once he slipped that blocksize limit into the core client, it would be there forever,
I think the interim limits of 250KB and 500KB invalidate your assertion.
I assume you're talking about the default, "soft block limits" that miners were free to choose, as long as they were under the 1MB hard (consensus) limit?
The raw data is also a source of some interesting discoveries. Individual pool operators sometimes try to optimize things in interesting (and quite different) ways. Here are some examples for some of the larger pools:
Antpool never adopted the 350k bytes value and jumped from 250k to 750k in June 2014.
BTCGuild adoped 475k bytes in March 2013, then 500k bytes from August 2013.
The now defunct DeepBit pool never went over 100k bytes.
Discus Fish (F2Pool) switched from 250k bytes to 32k in December 2013. After 6 weeks they variously tried 48k and 64k sizes for a few weeks each before adopting 916k in March 2014, then 933k in August 2014. In October 2014 they systematically started mining at 1M bytes; as such they consistently mine the largest blocks of all pools. Discus Fish also, curiously, mined a very large number of 0 transaction blocks for during the middle of 2014, suggesting some other peculiarities in their block selection.
Until November 2013 Eligius appears to have tried many quite large block sizes, before adopting a value just over 900k bytes.
GHash.IO ran with the 750k default for 2 weeks in March 2014 before moving back to 350k until late June. Their switch to larger blocks coincided with their decline in peak hash rate from a mean of 40% of the network. Curiously, both times they moved to 750k there was a corresponding large drop (about one sixth) in their share of the network; this could be complete coincidence but perhaps not!
Megabigpower has highly erratic maximum block sizes; the variation is seemingly far too large for their share of the network, with smaller pools showing dramatically more stability. This perhaps implies that they are usign some sort of custom maximum block size estimation software.
Most of the "unknown" hashing seems to be done by miners using the default sizes.
http://hashingit.com/analysis/39-the-myth-of-the-megabyte-bitcoin-blockEdit: I also recommend reading Shelby's latest blog post:
https://medium.com/@shelby_78386/apparently-not-400244cca3cc